Neuronal Networks in Food Science

How can AI help the food industry? What are some of the technologies developed by AI that can help formulators worldwide? 

Introduction

Artificial Neural Networks (ANNs) are computational models based on the biological functioning of the human brain. This computational model is employed in a wide range of applications for predicting parameters based on multiple input parameters with nonlinear behaviors among them. In the food science field, neural networks present a revolutionary tool for determining food properties, estimating product shelf life, assessing food quality, and various other applications, all at a reduced cost and execution time. Additionally, they offer the advantage of significantly reducing the use of reagents for physicochemical analysis.

Supervised and Unsupervised Training 

One of the most relevant properties of the neural network model is its learning capability, meaning its capacity to generate a response as close as possible to the expected parameter. Within the field of neural networks, there are two types of training:

• Supervised Training: This training is based on processing input information while knowing the expected output parameter. The network modifies the connection weights to reduce the difference between the expected and predicted parameters.

• Unsupervised Training: This training is based on processing input information without knowing the expected output value. In these conditions, the network identifies patterns and/or differences between the data without using an external output parameter.

In the field of food science, artificial neural networks offer a great opportunity for determining properties, shelf life, and quality of food products, among other applications. One of the advantages of using ANNs compared to traditional physicochemical analysis methods is the reduction in costs, analysis times, the use of harmful reagents, and the generation of chemical waste that requires proper disposal.

Several studies, such as those conducted by Sofu and Ekinci (2007) and Goyal and Goyal (2012, 2013a, 2013b), have shown that using neural networks for predicting the shelf life of dairy products such as cheese and yogurt generates results with high coefficients of determination, leading to significant reductions in costs and analysis times.

Another commonly studied application in food science is the monitoring of product quality by introducing various input parameters such as pH, color, fatty acid profile, hardness, among others. The use of ANNs with easily and rapidly measurable input variables during processing facilitates monitoring and reduces the reaction time to possible deviations.

In the area of cocoa and chocolate, various studies related to the quality of the initial cocoa and the final product have been conducted. For instance, León-Roque et al. (2016) analyzed the use of artificial neural networks for predicting the fermentation index of cocoa beans based on color measurement, where RGB values of the surface and extract of the beans were used as input parameters for the network. This study demonstrated that ANNs could allow the determination of the fermentation index of cocoa beans based on parameters such as the color of the exterior and the extract of the cocoa bean.

Another study in the chocolate industry, conducted by Sanja Podunavac-Kuzmanović et al. (2015), explored the determination of heavy metal content (Cu, Ni, Pb, and Al) in commercial chocolates. The results of this study indicate the possibility of predicting the metal content in different varieties of chocolates and defining the relationship between the different metal contents.

Conclusion

In conclusion, artificial neural networks are a promising alternative to replace traditional quality analysis methods, but they should be backed up by verified results obtained from said methods to guarantee that the training data used truly represents the studied or analyzed process. Successful application of artificial intelligence tools in the food industry may reduce costs of analysis and improve result times, thus improving production costs associated with material retention in facilities. 

References 

1. Millán – Trujillo, Felix (2022), “ Base teórica de las redes neuronales artificiales”. PB 7476: Tópico Experimental en Redes Neuronales en Ciencia de Los Alimentos, Universidad Simón Bolívar, Caracas, 1-7 de febrero del 2022. 
2. Goyal, S. (2013). Artificial neural networks (ANNs) in food science–A review. International Journal of Scientific World, 1(2), 19-28.
3. Sofu, A., and Ekinci, F.Y. (2007). Estimation of Storage Time of Yogurt with Artificial Neural Network Modeling. Journal of Dairy Science, 90(7), 3118–3125.
4. Cruz. A.G., Walter, E.H.M., Cadena, R.S., Faria, J.A.F., Bolini, H.M.A., and Fileti, A.M.F. (2009). Monitoring the authenticity of low-fat yogurts by an artificial neural network. Journal of Dairy Science, 92(10), 4797–4804. 
5. Goyal, Sumit, and Goyal, G.K. (2013). Intelligent artificial neural network computing models for predicting shelf life of processed cheese. Intelligent Decision Technologies, 7(2), 107-111.
6. Goyal, G.K., and Goyal, Sumit (2013). Cascade artificial neural network models for predicting shelf life of processed cheese. Journal of Advances in Information Technology, 4(2), 80-83.
7. Goyal, Sumit, and Goyal, G.K. (2012).Application of simulated neural networks as non-linear modular modeling method for predicting shelf life of processed cheese. Jurnal Intelek, 7(2), 48-54. 
8. León-Roque, N., Abderrahim, M., Nuñez-Alejos, L., Arribas, S. M., & Condezo-Hoyos, L. (2016). Prediction of fermentation index of cocoa beans (Theobroma cacao L.) based on color measurement and artificial neural networks. Talanta, 161, 31-39.
9. Jevrić, L., Podunavac-Kuzmanović, S., Švarc-Gajić, J., Kovačević, S., Vasiljević, I., Kecojević, I., & Ivanović, E. (2014). Artificial neural network approach to modelling of metal contents in different types of chocolates. Acta Chimica Slovenica, 62(1), 190-195.